Two things are supposed to go wrong with a morph effect like this. The deformation breaks mid-way and the join shows, and the figure that comes out the other side has nothing to do with the one that went in. Two clips were generated from one character to see which of the two actually turned up.
One of them did not happen at all. The other happened, once, in a place worth knowing about.
The split: 1280 by 720, 24 frames per second, 5.04 seconds. Shown here as frames in sequence; the full video and a GIF version are supplied alongside this document.
One character, plain background, limbs clear of the body
The starting character, generated for this article. Arms held out from the torso so the silhouette has no overlaps in it.
The source has three useful qualities: the whole body is in frame, the background is plain, and the limbs do not cross the torso. That clean silhouette makes the requested deformation easier to describe and makes the resulting shape changes easier to evaluate.
It is also deliberately a cartoon rather than a person. Stretching and splitting a photograph of somebody real is not something to do to a face that belongs to anyone, and the stylised version reads as a cartoon gag rather than as an injury.
What was asked for, and what the two morphs did
Stretch. The character stretches upward like elastic, the torso and legs drawing out long and thin until the head nearly reaches the top of the frame, then springs back down and stands whole again exactly as it began. The background stays flat and empty and nothing else in the frame moves. Cartoon 3D animation, no blood, no injury and no medical setting. Locked-off tripod shot, no zoom, no push in, no dolly, no pan and no crop, and the last frame has the same framing as the first frame.
Split. The character separates cleanly down the middle into a left half and a right half, the two halves slide apart sideways leaving a gap between them, hold for a moment, then slide back together and the character is whole again exactly as it began. …
The same four moments in each clip.
Neither did quite what the words said. The stretch elongated so far that the character left the frame — the silhouette is jammed against the top edge in 56 of the 121 frames, which is 2.3 seconds of a five-second clip with the head cropped off. And the split did not produce two halves: it produced two complete characters, standing apart with a gap of up to 411 pixels between them, for 74 of the 121 frames.
The split result is a reinterpretation rather than the requested left and right halves: the generator created two complete figures. To make the intended objects more explicit, describe each half separately, such as a left half with one arm and one boot, instead of describing only the act of separating.
The figure came back, in both clips, without being asked twice
The second predicted failure — the character not matching itself afterwards — did not occur. Comparing each frame's silhouette against the first frame's, the stretch ends at 0.918 of its starting shape and the split ends at 0.924. Both go a long way out first: the stretch bottoms out at 0.544, and the split reaches 0.026, which is very nearly no shape in common at all.
1.00 is the same shape as the opening frame. The split spends two and a half seconds at almost nothing in common and still comes home.
The result matters for workflow planning. In these two clips, a single five-second request included the starting pose, deformation, and return, so a separate three-part build was unnecessary for these particular outputs. Treat that as a Seedance 2.0 Mini observation rather than a guarantee, and test a short draft if the return frame is important.
The seam is on the way back, not on the way out
In these clips, the median frame-to-frame silhouette change falls between 0.001 and 0.003. Each clip also contains one much larger jump: 0.076 at 4.58 seconds for the stretch and 0.245 at 4.42 seconds for the split. The 0.245 jump is about 82 to 245 times the 0.001 to 0.003 reference range, depending on which end is used.
Both large jumps occur in the last second at reassembly. Separation is smooth in the two tested clips, while the return contains a visible snap. This small test does not prove that every run will break in the same place, but reassembly is the first place to inspect when reviewing another result.
What the two morphs cost
The cost lines as shown before each request, read on 15 September 2026.
The two effects used 51 credits in this session. Generating separate normal, deformation, and reassembly clips would have required three 5-second requests, or 75 credits, before editing under the same settings. That comparison assumes each segment needs a separate generation. The character and clips were made through the AI image generator and AI video generator flows. These figures apply to the shown configurations, account, and date. The credit balance is not used as a cross-check because it increased during the session despite the recorded spending, indicating an unobserved credit addition.
Put the transition on the frame it already breaks
In the split clip, the deformation remains smooth for about two and a half seconds before the largest reassembly jump. Inspect that frame, then try a cut, a brief transition, or a hold on a nearby clean frame. Preview the result afterward: a transition can soften the discontinuity, but it does not guarantee a seamless repair.
If reassembly itself needs to be visible, generate that action as a separate shot with slower, explicit movement and cut to it. In these two clips, the outward deformation was smoother than the return.
The character was generated from text; no photograph of a person was used, and body deformation is not something to apply to a photograph of someone else. Keep it to a character you made or to yourself.
Use a neutral, non-medical setting for this gag. Hospital clothing or emergency imagery can make a cartoon deformation look like illness or injury, changing the tone without improving the effect.
Rapid flashing can trigger seizures or other physical reactions in some viewers. Follow the W3C's three-flashes guidance, reduce repeated flashes, and provide a warning or static alternative when needed.
Also not covered: joining several segments with transitions, speed ramping for rhythm, audio, and export above 720p.
Written 15 September 2026. The character and both clips were generated in CapCut for this article; no real person appears in them. Both clips are 1280 by 720, 24 frames per second, 121 frames, 5.04 seconds. The silhouette was taken by measuring each frame's distance from the background color sampled at the four corners, then dropping components below 1,500 pixels; overlap is intersection over union against the first frame's silhouette. Figure counting was done in a horizontal band above the floor contact, because the shadow beneath the feet otherwise bridges two separate figures into one. An earlier attempt to measure the stretch by silhouette height was discarded: the character grows past the top of the frame, so the bounding box is clipped at 720 pixels and reports a 1.05x stretch for a deformation that plainly exceeds it. A frame-to-frame height-step threshold was also discarded, because the median step is zero and the threshold collapsed to zero with it. These findings apply to Seedance 2.0 Mini in the tested configuration; other models, settings, or later versions may produce different results.